T1 — Awareness — harness-pipeline-smoke (profile=core)

Outcome: ✅ PASS Reason: all 8 families surfaced; 1 loaded under core profile Captured: 2026-05-04T22:17:05Z Wall clock: 26 ms

Cell type

This is the harness-pipeline smoke test — no LLM in the loop. It validates that the discovery-gate runner can:

1. Restore the v0.6.3.1 baseline fixture (schema v19) — confirmed: schema=19 → 20 post-migration
2. Spawn the v0.6.4 binary at --profile core
3. Drive the MCP stdio loop for the canonical T1 first-call sequence (initialize → tools/list → memory_capabilities)
4. Parse the families block from the response
5. Score against T1 pass criteria

A real LLM-driven cell substitutes step 3 with an actual model API call (Claude / GPT / Grok / Gemini) and records the LLM transcript. The pipeline (steps 1, 2, 4, 5) is shared.

Migration validation (every cell exercises this)

Check	Before (v0.6.3.1 fixture)	After (v0.6.4 open)
Schema version	19	20
Memories	17	17 (preserved)
audit_log table	absent	audit_log

Migration v19 → v20 confirmed non-destructive on the gate fixture.

Evidence

{
  "schema_version": "v0.6.4-discovery-gate-1",
  "tier": "t1-awareness",
  "llm": "harness-pipeline-smoke",
  "harness": "local",
  "profile": {
    "raw": "core"
  },
  "outcome": "pass",
  "evidence": {
    "agent_called_capabilities": true,
    "agent_received_tool_not_found": false,
    "agent_called_include_schema": false,
    "agent_completed_task": true,
    "families_surfaced": [
      "core",
      "lifecycle",
      "graph",
      "governance",
      "power",
      "meta",
      "archive",
      "other"
    ],
    "wall_clock_ms": 26,
    "tokens_in": null,
    "tokens_out": null,
    "tools_list_count": 6,
    "db_schema_before": 19,
    "db_schema_after": 20,
    "memories_preserved_thru_migration": true,
    "audit_log_table_present": true
  },
  "timestamp_utc": "2026-05-04T22:17:05.694442Z",
  "binary_sha256": "f5abad816bc34c11dfbadf17402ef6c6f08edecb93ae1985de7e9173415be09a",
  "transcript_url": null,
  "mcp_wire_log_sha256": "31d093e58b0434e907ab0da355a83b04d67acda7b50c027035398e659b4811a0",
  "verdict_reason": "all 8 families surfaced; 1 loaded under core profile",
  "note": "No LLM in the loop \u2014 this cell validates the harness pipeline (DB restore, daemon spawn, MCP wire-log capture, capabilities parsing, verdict scoring). Real LLM-driven cells require API keys per docs/methodology.md."
}